public inbox for gcc-patches@gcc.gnu.org
 help / color / mirror / Atom feed
* [COMMITTED] Set nonzero bits for multiplication and divisions by a power of 2.
@ 2022-11-04 14:19 Aldy Hernandez
  0 siblings, 0 replies; only message in thread
From: Aldy Hernandez @ 2022-11-04 14:19 UTC (permalink / raw)
  To: GCC patches; +Cc: Andrew MacLeod, Aldy Hernandez

We're missing a lot of TLC in keeping track of nonzero bits across
range-ops.  It isn't an oversight, but just limited amount of hours to
implement stuff.

This patch keeps better track of the nonzero mask (really
maybe_nonzero bits as discussed) across multiplication and division
when the RHS is a power of 2.

It fixes PR107342 and also touches on PR55157.  In the latter, the
nonzero mask is being set quite late (CCP2) but could be set by evrp
time if we enhanced range-ops.  I have added tests from both PRs.

Tested

	PR tree-optimization/107342

gcc/ChangeLog:

	* range-op.cc (operator_mult::fold_range): New.
	(operator_div::fold_range): New.

gcc/testsuite/ChangeLog:

	* gcc.dg/tree-ssa/vrp122.c: New test.
	* gcc.dg/tree-ssa/vrp123.c: New test.
---
 gcc/range-op.cc                        | 59 ++++++++++++++++++++++++++
 gcc/testsuite/gcc.dg/tree-ssa/vrp122.c | 19 +++++++++
 gcc/testsuite/gcc.dg/tree-ssa/vrp123.c | 18 ++++++++
 3 files changed, 96 insertions(+)
 create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/vrp122.c
 create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/vrp123.c

diff --git a/gcc/range-op.cc b/gcc/range-op.cc
index 49ee7be3d3b..25c004d8287 100644
--- a/gcc/range-op.cc
+++ b/gcc/range-op.cc
@@ -1742,9 +1742,13 @@ cross_product_operator::wi_cross_product (irange &r, tree type,
 
 class operator_mult : public cross_product_operator
 {
+  using range_operator::fold_range;
   using range_operator::op1_range;
   using range_operator::op2_range;
 public:
+  virtual bool fold_range (irange &r, tree type,
+			   const irange &lh, const irange &rh,
+			   relation_trio = TRIO_VARYING) const final override;
   virtual void wi_fold (irange &r, tree type,
 		        const wide_int &lh_lb,
 		        const wide_int &lh_ub,
@@ -1762,6 +1766,32 @@ public:
 			  relation_trio) const;
 } op_mult;
 
+bool
+operator_mult::fold_range (irange &r, tree type,
+			   const irange &lh, const irange &rh,
+			   relation_trio trio) const
+{
+  if (!cross_product_operator::fold_range (r, type, lh, rh, trio))
+    return false;
+
+  if (lh.undefined_p ())
+    return true;
+
+  tree t;
+  if (rh.singleton_p (&t))
+    {
+      wide_int w = wi::to_wide (t);
+      int shift = wi::exact_log2 (w);
+      if (shift != -1)
+	{
+	  wide_int nz = lh.get_nonzero_bits ();
+	  nz = wi::lshift (nz, shift);
+	  r.set_nonzero_bits (nz);
+	}
+    }
+  return true;
+}
+
 bool
 operator_mult::op1_range (irange &r, tree type,
 			  const irange &lhs, const irange &op2,
@@ -1902,10 +1932,39 @@ public:
 		        const wide_int &rh_ub) const;
   virtual bool wi_op_overflows (wide_int &res, tree type,
 				const wide_int &, const wide_int &) const;
+  virtual bool fold_range (irange &r, tree type,
+			   const irange &lh, const irange &rh,
+			   relation_trio trio) const final override;
 private:
   enum tree_code code;
 };
 
+bool
+operator_div::fold_range (irange &r, tree type,
+			  const irange &lh, const irange &rh,
+			  relation_trio trio) const
+{
+  if (!cross_product_operator::fold_range (r, type, lh, rh, trio))
+    return false;
+
+  if (lh.undefined_p ())
+    return true;
+
+  tree t;
+  if (rh.singleton_p (&t))
+    {
+      wide_int wi = wi::to_wide (t);
+      int shift = wi::exact_log2 (wi);
+      if (shift != -1)
+	{
+	  wide_int nz = lh.get_nonzero_bits ();
+	  nz = wi::rshift (nz, shift, TYPE_SIGN (type));
+	  r.set_nonzero_bits (nz);
+	}
+    }
+  return true;
+}
+
 bool
 operator_div::wi_op_overflows (wide_int &res, tree type,
 			       const wide_int &w0, const wide_int &w1) const
diff --git a/gcc/testsuite/gcc.dg/tree-ssa/vrp122.c b/gcc/testsuite/gcc.dg/tree-ssa/vrp122.c
new file mode 100644
index 00000000000..b2ddcda023c
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/tree-ssa/vrp122.c
@@ -0,0 +1,19 @@
+// { dg-do compile }
+// { dg-options "-O2 -fdump-tree-evrp-details" }
+
+void gg(void);
+int f(unsigned t)
+{
+  unsigned g = t*16;
+  if (g==0)  return 1;
+  gg();
+  gg();
+  gg();
+  gg();
+  gg();
+  gg();
+  if (g<=4)  return 1;
+  return 0;
+}
+
+// { dg-final { scan-tree-dump "Global Exported: g_.* NONZERO 0x.*fff0" "evrp" } }
diff --git a/gcc/testsuite/gcc.dg/tree-ssa/vrp123.c b/gcc/testsuite/gcc.dg/tree-ssa/vrp123.c
new file mode 100644
index 00000000000..1ad3caa4384
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/tree-ssa/vrp123.c
@@ -0,0 +1,18 @@
+// { dg-options "-O1 -fdump-tree-dom3-raw" }
+
+extern int
+__attribute__((const))
+foo4b (int);
+
+int f4b (unsigned int r)
+{
+  if (foo4b (r))
+    r *= 8U;
+
+  if ((r / 2U) & 2U)
+    r += foo4b (r);
+
+  return r;
+}
+
+// { dg-final { scan-tree-dump-times {gimple_call <foo4b,} 1 dom3 } }
-- 
2.38.1


^ permalink raw reply	[flat|nested] only message in thread

only message in thread, other threads:[~2022-11-04 14:19 UTC | newest]

Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2022-11-04 14:19 [COMMITTED] Set nonzero bits for multiplication and divisions by a power of 2 Aldy Hernandez

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).